AI and the Enclosure of Science's Shared Literature
A new arXiv paper frames AI's impact on science as a threat to the information commons of the literature.
Research Digest
Preprints and papers, rewritten in plain English.
Inspectable AI Participants Help Rehearse Qualitative Interviews
A new arXiv preprint proposes making LLM-simulated interview participants inspectable, so researchers can rehearse qualitative interviews with greater transparency.
Understanding Human Oversight in Agentic AI
A new arXiv paper examines how human oversight changes when AI agents take on complex tasks.
Chronic Illness Storytelling: AI and Communities Aid Narrative Translation
A new arXiv paper examines how people with chronic illness translate bodily experiences into narratives with help from AI and community support.
Agentic AI's Micro-Waits: Too Long to Watch, Too Short to Leave
A new study examines the awkward idle time users face while AI agents work for minutes at a stretch, and suggests design directions for these micro-waits.
Why Social Companion AI Appeal Is More Than Features
A new paper argues that the appeal of social companion AI cannot be reduced to its positive traits like being nonjudgemental and always available.
Designing AI memory with 'negative space' for forgetting
A new arXiv paper proposes a governance framework for AI memory that prioritizes forgetting disclosed information over future personalization benefits.
Guided Diffusion Aims to Make Synthetic Text Safer
A new arXiv paper proposes guided diffusion for generating synthetic text that could reduce privacy risks from LLM training data.
WhatIf: Interactive tool for policy reasoning with LLM agent simulations
A new arXiv paper introduces WhatIf, an interactive system for exploring large-scale LLM agent simulations to support policy decisions under uncertainty.
Flow Control Steers Vision-Language-Action Models in Real Time
A new arXiv paper introduces a simple method for using generic inputs, such as a keyboard, to steer vision-language-action models in real time.
Embodied VLMs' Social Reasoning Under Scrutiny in New arXiv Paper
A new arXiv paper asks whether embodiment-specialized vision-language models retain the social reasoning abilities of general-purpose models when used as reasoning layers for robot teams.
AI Agents Quietly Make Design Decisions for Developers
A new framework called Aporia highlights how AI programming agents can silently take over design decisions, raising questions about developer control.
Chatbot-Based Assistants Proposed for Cognitive Decline Screening
A new preprint suggests that conversational AI could make early cognitive decline screening more scalable for older adults.
How Korean-English Code-Switching Works in LLM Speaking Practice
New research examines code-switching in AI-driven language learning, a context NLP studies have largely overlooked.
TracePrism partitions event logs by sub-process for better process mining
A new arXiv paper proposes iterative, encoding-aware trace partitioning to handle event logs that mix different sub-processes.
WeGR Paper Proposes Behavior-Aware Tokenization for Industrial Search
A new arXiv paper introduces WeGR, a generative retrieval method that incorporates user behavior into tokenization and aligns preferences to improve industrial search.
DeclineCraft: Benchmarking When RAG Should Abstain
A new benchmark measures whether retrieval-augmented models know when to decline answering.
A Publisher-Controlled Web Architecture for AI Agents
A new paper argues that the web, built for human browsing, needs a native publishing model for AI consumers.